[adrisabate]

Hello SU2 community,

In short, my question is: can anybody confirm me that the Rotating Mesh option in SU2 version 6.0.1 (or version 5) works in combination with the RANS solver?

I am trying to reproduce the Caradonna & Tung experiments on SU2 version 6 for my master thesis work; I have been able to perform Euler simulations but I am not able to arrive to a converged solution in steady RANS, or if I do it does not look physical. It this a bug of SU2 version 6 ?

I know about the paper AIAA 2014 SU2: Open-Source Analysis and Design Technology for Turbulent Flows; the Caradonna & Tung test in this paper is my reference for the configuration parameters and physical variables. But as it is from 2014 the version used must have been a release of SU2 version 3.

I also remarked in this forum there have been issues in the past with the rotating mesh option after a new release modifies of the code.



(From now on at this post I'm just adding more details to this question)

There is this exception to convergence issues:
If at some configuration the solution does not diverge, I can see random mesh points on the surface with large density (picture attached, in my office we call it the "leopard").

The parameter that allows me to have residual and force coefficient convergence is reducing the resolution/restriction of the linear solver:

% ------------------------ LINEAR SOLVER DEFINITION ---------------------------%
%
% Linear solver or smoother for implicit formulations (BCGSTAB, FGMRES, SMOOTHER_JACOBI, 
%                                                      SMOOTHER_ILU, SMOOTHER_LUSGS, 
%                                                      SMOOTHER_LINELET)
LINEAR_SOLVER= FGMRES
%
% Preconditioner of the Krylov linear solver (ILU, LU_SGS, LINELET, JACOBI)
LINEAR_SOLVER_PREC= LU_SGS
%
% Minimum error of the linear solver for implicit formulations
LINEAR_SOLVER_ERROR= 1E-4 % Default is 1E-6
%
% Max number of iterations of the linear solver for the implicit formulation
LINEAR_SOLVER_ITER= 4 % Default is 20

About my specific case (also share with you my preferred CFG file):

• I use ANSYS Meshing to create an unstructured mesh in CGNS, with automatically generated prisms representing the boundary layer and tetrahedrons for the rest of the fluid. My geometry has no hub and a thick trailing edge where I try to have 3 points plus the edges. I started with the 8 degree collective pitch case and later reduced it to 5 degrees.
• I have tried different refinement levels and boundary layers, suspecting my mesh wasn’t good enough.
• I have tried both constant temperature wall and adiabatic wall boundary conditions. I have tried to divide my farfield boundaries in different markers.
• Then I tried a wide range of solver parameter combinations, starting from the solver parameters used in the paper AIAA 2014, then trying to change them to favor stability, but also tested some riskier (in my opinion less stable) parameters.

In all cases, sooner or later (depending on if I use 2 multi-grid levels or not, or on the limits I apply to the variable CFL) I arrive to a “SU2 has diverged” message, or eventually I stop them if I see a “There is X cells with non-physical values” message and I confirm in the solution my density or temperature is not the order of magnitude it should be. With the only exception of when I don't let the solver iterate of a given timestep.

However, I don’t feel this should be this difficult: the geometry isn’t the most complex, meshing shouldn’t be a challenge, Mach number is low...


Any commentary is appreciated, especially if anybody could confirm me they are obtaining results on 3D Rotating Mesh + RANS steady simulations. If it is not the case, inform the development team!

In any case, it is already late to be included in my master thesis but it would be nice to know what happened.


Cheers,